/*!
* \internal
* \brief Handle cluster communication related to pacemaker_remote node leaving
*
* \param[in] node_name Name of lost node
*/
static void
remote_node_down(const char *node_name)
{
xmlNode *update;
int call_id = 0;
int call_opt = crmd_cib_smart_opt();
crm_node_t *node;
/* Clear all node attributes */
update_attrd_remote_node_removed(node_name, NULL);
/* Ensure node is in the remote peer cache with lost state */
node = crm_remote_peer_get(node_name);
CRM_CHECK(node != NULL, return);
crm_update_peer_state(__FUNCTION__, node, CRM_NODE_LOST, 0);
/* Notify DC */
send_remote_state_message(node_name, FALSE);
/* Update CIB node state */
update = create_xml_node(NULL, XML_CIB_TAG_STATUS);
do_update_node_cib(node, node_update_cluster, update, __FUNCTION__);
fsa_cib_update(XML_CIB_TAG_STATUS, update, call_opt, call_id, NULL);
if (call_id < 0) {
crm_perror(LOG_ERR, "%s CIB node state update", node_name);
}
free_xml(update);
}
/*!
* \internal
* \brief Handle effects of a remote RA command on node state
*
* \param[in] cmd Completed remote RA command
*/
static void
check_remote_node_state(remote_ra_cmd_t *cmd)
{
/* Only successful actions can change node state */
if (cmd->rc != PCMK_OCF_OK) {
return;
}
if (safe_str_eq(cmd->action, "start")) {
remote_node_up(cmd->rsc_id);
} else if (safe_str_eq(cmd->action, "migrate_from")) {
/* After a successful migration, we don't need to do remote_node_up()
* because the DC already knows the node is up, and we don't want to
* clear LRM history etc. We do need to add the remote node to this
* host's remote peer cache, because (unless it happens to be DC)
* it hasn't been tracking the remote node, and other code relies on
* the cache to distinguish remote nodes from unseen cluster nodes.
*/
crm_node_t *node = crm_remote_peer_get(cmd->rsc_id);
CRM_CHECK(node != NULL, return);
crm_update_peer_state(__FUNCTION__, node, CRM_NODE_MEMBER, 0);
} else if (safe_str_eq(cmd->action, "stop")) {
/*!
* \internal
* \brief Return host's hash table entry (creating one if needed)
*
* \param[in] values Hash table of values
* \param[in] host Name of peer to look up
* \param[in] xml XML describing the attribute
*
* \return Pointer to new or existing hash table entry
*/
static attribute_value_t *
attrd_lookup_or_create_value(GHashTable *values, const char *host, xmlNode *xml)
{
attribute_value_t *v = g_hash_table_lookup(values, host);
int is_remote = 0;
crm_element_value_int(xml, F_ATTRD_IS_REMOTE, &is_remote);
if (is_remote) {
/* If we previously assumed this node was an unseen cluster node,
* remove its entry from the cluster peer cache.
*/
crm_node_t *dup = crm_find_peer(0, host);
if (dup && (dup->uuid == NULL)) {
reap_crm_member(0, host);
}
/* Ensure this host is in the remote peer cache */
CRM_ASSERT(crm_remote_peer_get(host) != NULL);
}
if (v == NULL) {
v = calloc(1, sizeof(attribute_value_t));
CRM_ASSERT(v != NULL);
v->nodename = strdup(host);
CRM_ASSERT(v->nodename != NULL);
v->is_remote = is_remote;
g_hash_table_replace(values, v->nodename, v);
}
return(v);
}
/*!
* \internal
* \brief Handle cluster communication related to pacemaker_remote node leaving
*
* \param[in] node_name Name of lost node
*/
static void
remote_node_down(const char *node_name)
{
xmlNode *update;
int call_id = 0;
int call_opt = crmd_cib_smart_opt();
crm_node_t *node;
/* Purge node from attrd's memory */
update_attrd_remote_node_removed(node_name, NULL);
/* Purge node's operation history and transient attributes from CIB */
erase_status_tag(node_name, XML_CIB_TAG_LRM, call_opt);
erase_status_tag(node_name, XML_TAG_TRANSIENT_NODEATTRS, call_opt);
/* Ensure node is in the remote peer cache with lost state */
node = crm_remote_peer_get(node_name);
CRM_CHECK(node != NULL, return);
crm_update_peer_state(__FUNCTION__, node, CRM_NODE_LOST, 0);
/* Notify DC */
send_remote_state_message(node_name, FALSE);
/* Update CIB node state */
update = create_xml_node(NULL, XML_CIB_TAG_STATUS);
do_update_node_cib(node, node_update_cluster, update, __FUNCTION__);
fsa_cib_update(XML_CIB_TAG_STATUS, update, call_opt, call_id, NULL);
if (call_id < 0) {
crm_perror(LOG_ERR, "%s CIB node state update", node_name);
}
free_xml(update);
}
/*!
* \internal
* \brief Handle cluster communication related to pacemaker_remote node joining
*
* \param[in] node_name Name of newly integrated pacemaker_remote node
*/
static void
remote_node_up(const char *node_name)
{
int call_opt, call_id = 0;
xmlNode *update, *state;
crm_node_t *node;
CRM_CHECK(node_name != NULL, return);
crm_info("Announcing pacemaker_remote node %s", node_name);
/* Clear node's operation history. The node's transient attributes should
* and normally will be cleared when the node leaves, but since remote node
* state has a number of corner cases, clear them here as well, to be sure.
*/
call_opt = crmd_cib_smart_opt();
erase_status_tag(node_name, XML_CIB_TAG_LRM, call_opt);
erase_status_tag(node_name, XML_TAG_TRANSIENT_NODEATTRS, call_opt);
/* Clear node's probed attribute */
update_attrd(node_name, CRM_OP_PROBED, NULL, NULL, TRUE);
/* Ensure node is in the remote peer cache with member status */
node = crm_remote_peer_get(node_name);
CRM_CHECK(node != NULL, return);
crm_update_peer_state(__FUNCTION__, node, CRM_NODE_MEMBER, 0);
/* pacemaker_remote nodes don't participate in the membership layer,
* so cluster nodes don't automatically get notified when they come and go.
* We send a cluster message to the DC, and update the CIB node state entry,
* so the DC will get it sooner (via message) or later (via CIB refresh),
* and any other interested parties can query the CIB.
*/
send_remote_state_message(node_name, TRUE);
update = create_xml_node(NULL, XML_CIB_TAG_STATUS);
state = create_node_state_update(node, node_update_cluster, update,
__FUNCTION__);
/* Clear the XML_NODE_IS_FENCED flag in the node state. If the node ever
* needs to be fenced, this flag will allow various actions to determine
* whether the fencing has happened yet.
*/
crm_xml_add(state, XML_NODE_IS_FENCED, "0");
/* TODO: If the remote connection drops, and this (async) CIB update either
* failed or has not yet completed, later actions could mistakenly think the
* node has already been fenced (if the XML_NODE_IS_FENCED attribute was
* previously set, because it won't have been cleared). This could prevent
* actual fencing or allow recurring monitor failures to be cleared too
* soon. Ideally, we wouldn't rely on the CIB for the fenced status.
*/
fsa_cib_update(XML_CIB_TAG_STATUS, update, call_opt, call_id, NULL);
if (call_id < 0) {
crm_perror(LOG_WARNING, "%s CIB node state setup", node_name);
}
free_xml(update);
}
/*!
* \brief Handle a CRM_OP_REMOTE_STATE message by updating remote peer cache
*
* \param[in] msg Message XML
*
* \return Next FSA input
*/
static enum crmd_fsa_input
handle_remote_state(xmlNode *msg)
{
const char *remote_uname = ID(msg);
const char *remote_is_up = crm_element_value(msg, XML_NODE_IN_CLUSTER);
crm_node_t *remote_peer;
CRM_CHECK(remote_uname && remote_is_up, return I_NULL);
remote_peer = crm_remote_peer_get(remote_uname);
CRM_CHECK(remote_peer, return I_NULL);
crm_update_peer_state(__FUNCTION__, remote_peer,
crm_is_true(remote_is_up)?
CRM_NODE_MEMBER : CRM_NODE_LOST, 0);
return I_NULL;
}
/*!
* \internal
* \brief Handle cluster communication related to pacemaker_remote node leaving
*
* \param[in] node_name Name of lost node
* \param[in] opts Whether to keep or erase LRM history
*/
static void
remote_node_down(const char *node_name, const enum down_opts opts)
{
xmlNode *update;
int call_id = 0;
int call_opt = crmd_cib_smart_opt();
crm_node_t *node;
/* Purge node from attrd's memory */
update_attrd_remote_node_removed(node_name, NULL);
/* Purge node's transient attributes */
erase_status_tag(node_name, XML_TAG_TRANSIENT_NODEATTRS, call_opt);
/* Normally, the LRM operation history should be kept until the node comes
* back up. However, after a successful fence, we want to clear it, so we
* don't think resources are still running on the node.
*/
if (opts == DOWN_ERASE_LRM) {
erase_status_tag(node_name, XML_CIB_TAG_LRM, call_opt);
}
/* Ensure node is in the remote peer cache with lost state */
node = crm_remote_peer_get(node_name);
CRM_CHECK(node != NULL, return);
crm_update_peer_state(__FUNCTION__, node, CRM_NODE_LOST, 0);
/* Notify DC */
send_remote_state_message(node_name, FALSE);
/* Update CIB node state */
update = create_xml_node(NULL, XML_CIB_TAG_STATUS);
create_node_state_update(node, node_update_cluster, update, __FUNCTION__);
fsa_cib_update(XML_CIB_TAG_STATUS, update, call_opt, call_id, NULL);
if (call_id < 0) {
crm_perror(LOG_ERR, "%s CIB node state update", node_name);
}
free_xml(update);
}
